Podcast Episode
High-bandwidth memory is the specialised, ultra-fast memory that sits alongside AI accelerators such as Nvidia's GPUs, feeding them data quickly enough to keep up with the enormous demands of training and running large AI models. It is a critical bottleneck in modern AI infrastructure, and demand has exploded accordingly.
Samsung shipped its first commercial HBM4 units in February 2026, delivering a consistent processing speed of 11.7 gigabits per second, roughly 46 percent above the industry standard, with total memory bandwidth per stack reaching up to 3.3 terabytes per second. The company expects its overall HBM sales to more than triple in 2026 compared with the prior year, and analysts project HBM4 revenue could exceed $1.2 billion by the end of June and top $10 billion for the full year.
Samsung's HBM4 Chip Sales Top $1 Billion Just Four Months After Launch
June 24, 2026
0:00
5:12
Samsung Electronics has crossed $1 billion in revenue for its sixth-generation HBM4 memory chips, only four months after becoming the first company in the world to begin mass production in February 2026. The milestone marks a comeback for Samsung after it fell behind rival SK Hynix in the booming market for AI memory. Analysts expect Samsung's HBM4 revenue to surpass $10 billion for the full year.
Samsung Storms Back Into the AI Memory Race
Samsung Electronics has crossed the $1 billion revenue mark for its sixth-generation high-bandwidth memory chips, known as HBM4, just four months after becoming the first company in the world to begin mass production in February 2026. The figure, reported by industry sources via Yonhap News Agency, signals a dramatic turnaround for a company that had recently been trailing in one of the most lucrative corners of the semiconductor industry.High-bandwidth memory is the specialised, ultra-fast memory that sits alongside AI accelerators such as Nvidia's GPUs, feeding them data quickly enough to keep up with the enormous demands of training and running large AI models. It is a critical bottleneck in modern AI infrastructure, and demand has exploded accordingly.
A Race to Recover Lost Ground
Samsung's achievement comes after a difficult stretch. SK Hynix overtook Samsung in annual operating profits for the first time in 2025, buoyed by its dominance in HBM chips used in AI data centres. By the third quarter of 2025, SK Hynix held roughly 62 percent of the HBM market while Samsung trailed with about 17 percent, according to Counterpoint Research.Samsung shipped its first commercial HBM4 units in February 2026, delivering a consistent processing speed of 11.7 gigabits per second, roughly 46 percent above the industry standard, with total memory bandwidth per stack reaching up to 3.3 terabytes per second. The company expects its overall HBM sales to more than triple in 2026 compared with the prior year, and analysts project HBM4 revenue could exceed $1.2 billion by the end of June and top $10 billion for the full year.
Capacity Expansion Amid Surging Demand
Both Samsung and SK Hynix are scaling up production to meet relentless demand from AI infrastructure builders. Samsung is reportedly expanding its HBM production capacity by around 50 percent in 2026, while SK Hynix has announced plans to increase infrastructure investment by more than four times its previously disclosed figure. Bank of America estimates the total HBM market will reach $54.6 billion in 2026, a 58 percent increase year on year. Samsung is also preparing its next-generation HBM4E, with sampling expected in the second half of 2026 and custom HBM samples reaching customers in 2027.The Broader AI Memory Supercycle
The wider semiconductor industry continues to expand rapidly on the back of AI demand. Valued at over $670 billion in 2024, the global semiconductor market is projected to exceed $1 trillion by 2030, according to Yole Group, with the memory segment growing even faster and potentially surpassing $325 billion by 2030. Still, the competition remains fierce: UBS predicts SK Hynix will capture roughly 70 percent market share in HBM4 chips for Nvidia's next-generation Rubin platform. Samsung may have recaptured momentum, but the battle for AI memory supremacy is far from over.Published June 24, 2026 at 11:41am